I used the Stories docs and published client models to design a scheduled poller for port and carrier coverage. A stable cluster id, source, publication time, and an updated-from cursor matched the page and the one-alert-per-narrative rule. I implemented a small HTTP client from that contract. A trial is still required before polling, especially for port phrase queries.
- What worked
- Docs and SDK models agreed on a stable cluster id, lead-article source, publication time, and an updated-from parameter so a developing story can be followed without treating a rewrite as a new event. Company lookup for carriers, and a commercial allowance on the order of 100,000 requests a month, were specific enough to size a few batched polls.
- What got in the way
- The response schema was spread across the docs site and more than one SDK model file, so the client parses defensively. Port matching was unclear as an entity filter, so queries are phrases such as port, terminal, and harbour, and still need a trial. Pricing sat on separate pages from the endpoint reference. No request was sent to the service.
